The Future of U.S. Weather Modeling

Dr. Cliff Mass has produced another outstanding essay on the hoped-for future of U.S. meteorological computer modeling. It is here and it is highly recommended if you are interested in the topic.

I wish to pass along some miscellaneous thoughts of my own on this topic:

• Before we embark on an project to completely revamp U.S. meteorological computer modeling we need to ask ourselves the most vital question of all: how good is good enough? Since we know there will never be perfection, how good is good enough? We need objective yardsticks against which to measure our success. 

• How do we bring consistency into the process? The models currently crank out uncalibrated probabilities (bad!) and the quality of their forecasts varies hugely from storm to storm. 

We just had a snow-producing system move across Kansas and the models were awful. No consistency or skill. 

Here is the composite forecast for Wichita which received 1.4 inches. Forecast amounts were from zero to 9 inches. The average forecast was 4.5. The system over forecasted the snow amount. 

Here is the forecast for Dodge City which received 10 inches. Forecast amounts were from zero to 8 inches. The average was 3.8 inches. Dodge City's actual snow was both outside of the range of models' forecasts and had a lower average forecast than Wichita's in spite of Dodge City receiving 7 times more snow. 

In spite of model physics and computational power undreamed of twenty years ago, I can't tell that the global models have consistently improved at the time scales that are most important (zero to 60 hours). Yes, there is some aggregate improvement but meteorologists need consistency and specificity that we are not currently receiving. 

Note: There is improvement at 4-6 days and, perhaps, farther out in time. That is helpful but if we can't tell you the amount of snow the day before...